Cybernetic microbial detection system using transfer learning

• Published in: Multimedia tools and applications Vol. 79; no. 7-8; pp. 5225 - 5242
• Main Authors: Dinesh Jackson Samuel R, Rajesh Kanna B
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2020; Springer Nature B.V
• Subjects: Computer Communication Networks; Computer Science; Convolution; Cybernetics; Data Structures and Information Theory; Diagnosis; Error analysis; Image acquisition; Image processing; Infectious diseases; Medical imaging; Microorganisms; Microscopes; Microscopy; Multimedia Information Systems; Neural networks; Recognition; Remote regions; Screening; Smear; Special Purpose and Application-Based Systems; Support vector machines; Tuberculosis; TB bacilli detection; CNN-SVM; Motorized microscopic stage; Humanitarian nursing; Microbial recognition system
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-018-6356-z

Microorganisms are single or multi-cellular living organism viewed under a microscope. For pathological study, the images of these microbes are captured using microscopes and image processing is done for further analysis. The World Health Organization (WHO) recommends to view at least 300 Field-of-Views (FOV) manually, for diagnosing and reporting the level of infection. These operation involved for analysis requires skilled technician for error free results. When the number of images for diagnosis increases, it becomes cumbersome for the technicians as there is a chance of ambiguity in results, which hampers the sensitivity of the study. In this work, a Cybernetic Microbial Detection System (CMDS) has been proposed. As in the first stage, all field of views are acquired from the specimen using a programmable microscopic stage coupled with an acquisition system. Herein, the user defines a scanning pattern for the microscopic stage movement, which facilitates the data acquisition during specimen screening. In the second stage microbial recognition system is proposed, wherein a transfer learning technique is implemented by customizing Visual Geometry Group (VGG16/19) layered convolution neural network coupled to the Support Vector Machine (SVM). These modified stack layers (VGG + SVM) has been trained/tested with microscopy images of ZN stained Tuberculosis (TB) specimem obtained from a open source database [3] and also acquired microscopy images from several patients sputum smear specimen. The accuracy obtained from the TB recognition system is 83.404 and 86.6% for VGG16-SVM, VGG19-SVM respectively. Thus the proposed screening process reduces the reliance on skilled technicians and facilitates the humanitarian nursing. Hence, the proposed cybernetic system might be useful to human community in remote regions for rapid diagnosis and detection of infectious diseases.